How to handle many new population PK and other literature models

[billdenney]

@mattfidler and others, I'm in the process of building what will likely be hundreds of population PK models into the package. For each of them, I intend to have a vignette to show that they are working correctly.

This won't work well given typical infrastructure for surfacing vignettes. CRAN checks will likely fail due to taking too long to build the vignettes, and finding vignettes on how to use the package in general will be harder to find if there are many literature models to sort through.

I think that for the literature models, it would probably make sense to do a few things differently:

1. Put them into a GitHub pages-only location (e.g. vignettes/articles) which won't go into the main CRAN distribution.
2. Link them from the list of models vignette so that people can find them on the GitHub pages location.
3. Put them into a separate section in the pkgdown website so that there is one list of general how-to vignettes in pkgdown and another list of literature models.

Do you have any thoughts on this implementation?

[billdenney]

Another option could be that they live in a separate package, though I don't prefer that if we can find a good solution that keeps them here.

[mattfidler]

I think that they should be in the vignettes/articles making sure it is .Rbuildignore. and then link them from the list of models. You could have this generated by some keywords like disease area, model type, etc.

There are only a few issues here:

1. The index that is included in the actual vignette may be out of sync (mabye link to the vignette if it is updated with every new literature model).
2. The length of the model library may be increased such that it is too long to render in a github actions

Other than that, I think it would be better if they live in the same package.

There should be some discussion on how they are handled in nlmixr2shiny as well since many may want to interface from there instead 😄

[billdenney]

That makes sense to me.

We can start the model list vignette with a link to the GitHub pages vignette saying that additional models may be available on GitHub.

If we get to the point that we are taking longer than GitHub actions allow, we could consider something more complicated like parallel vignette building, or longer-term or pre-calculation of results.

[billdenney]

I implemented them in the articles directory with a .Rbuildignore, so it only shows up here and not on CRAN. It currently takes ~1.5 hours to build them all which is notably shorter than the 6-hour limit for runners. I also proposed a parallel patch to pkgdown (r-lib/pkgdown#2989) which should cut the pkgdown build by a factor of ~4, if accepted.